OpenAI has released GPT-5.6 Luna, a lighter model tier priced at 80 percent below the standard GPT-5.6 rate. The OpenAI API pricing page lists the new tier with sharply reduced input and output token costs, aimed at high-volume developers who found the flagship too expensive at scale.
The discount is the headline, but the placement matters more. Luna sits below the full GPT-5.6 release and is described as optimized for everyday tasks such as summarization, classification, extraction, and lightweight chat. It is not positioned as a replacement for the flagship, and OpenAI keeps the two tiers separate so teams can route requests by difficulty.
What the cut changes
An 80 percent reduction in token price changes the economics of several workloads at once. A product that had to cache aggressively or send shorter prompts can now afford longer context and more retries. For a small team, the tier can make a model-backed feature profitable that previously lost money on every call.
The tradeoff is in the ceiling. Luna scores below the full model on complex reasoning and long-horizon tasks, and it can be verbose when a shorter answer would do. Teams that route the easy traffic to Luna and the hard traffic to the flagship get the best of both: lower average cost without a visible quality drop on the work users actually see.
There is also a practical benefit to a separate tier: it gives pricing a dial. When a workload outgrows the cheap tier, the upgrade path is a config change rather than a rebuild. That is the kind of flexibility that lets a product start small and scale without rewriting its integration.
As with any price cut, the useful question is total cost, not the headline number. Luna is cheaper per token, but a model that produces more tokens to reach the same answer can erase part of the saving. Testing the tier on the real workload is the only way to know whether the discount is a bargain.
For now, the direction is clear. OpenAI is segmenting its lineup by cost as well as capability, and an 80 percent cut on a core tier is a signal that price competition in model APIs is not slowing down.
